Only compile files defined as source.

[kul]

Currently doing an aot will also try to compile data_readers.clj or any other resource file ending with .clj e.g. an edn file. This patch modifies aot and java compilation by limiting file set with :source mask only.

[micha]

Thanks for the PR! But won't this make it impossible to AOT Clojure namespaces in :resource-paths?

Tasks that compile or build things should always operate on files with input role. Whether a file is in :source-paths or :resource-paths is not relevant to a build task, because both add the input role. The difference between :source-paths and :resource-paths is that only :resource-paths has the output role. Packaging tasks, unlike build tasks, operate on files with the output role.

With this patch you couldn't, for example, build a JAR with AOT and the Clojure source files in the JAR, because the usual way to include things in packaging is to ensure that they have the output role (i.e. in this case put them in :resource-paths).

<rant> This, incidentally, is an example of how we abuse file extensions. Why not use the .edn extension for EDN files, leaving .clj for Clojure namespaces? It's for this reason that we used build.boot for example, and not build.clj. It takes 2 seconds to associate build.boot with file format for syntax highlighting in your editor, but in return we get useful information from the file extension and our tools can help us do things. </rant>

So in this case the only things we can do are:

• examine the file and use some heuristic to determine whether it looks like a compilable clojure namespace
• special-case handling for data_readers.clj

What do you think?

[kul]

Hi @micha ,
Totally agree with the rant. But I still do not quite agree on the part of compiling resources.

The problem you pointed out of both aot and source packaging is reproducible with the patch, I didnt test it enough. I was under the impression that adding to both resource and source paths would handle this case but it turns out the get-files function works differently and would not return a file for source mask if added to both src and resource.

I started this patch with a filter for data_readers but soon figured its brittleness. No matter what we argue people are at some point going to do weird things, data_readers.clj is a great example of such abuse.

What do you propose?

[micha]

@kul I propose one or both of the following:

• Examine files with .clj extension and use some rules for testing whether the file is a Clojure namespace or not (eg. look for ns macro)
• Add special handling for known non-namespace .clj files, data_readers.clj being one of them
  • We may want to add options to the aot task similar to what we've done with the uber task: options to allow regex exclusion with sane defaults.

[kul]

@micha Yes exclusion regex makes total sense. Ok i will have a crack at it and squash in new commits. Thanks!